Price Discovery, Volatility Spillovers and Adequacy of Spec - PowerPoint Presentation

Slide1

Price Discovery, Volatility Spillovers and Adequacy of Speculation in Cheese Spot and Futures Markets

Marin BozicUniversity of Minnesota-Twin CitiesNDSU Seminar, 10/28/2011

1Slide2

Motivation: Volatility in Dairy Sector

2Slide3

3

Motivation: How to Model Agricultural PricesSlide4

4

Motivation: How to Model Speculative Influence?Slide5

Volatility in the Dairy Sector: Why?

S

D

D

′

Quantity

Price

5Slide6

Volatility in the Dairy Sector: Why?

6Slide7

Dealing with High Volatility

Price Support ProgramsMilk Income Loss Contract

7

Catastrophic Insurance (LGM-Dairy)

Market-based instruments: Dairy Futures & Options, OTCs

Herd Termination Programs

Social Insurance

Supply ManagementSlide8

Purpose of this paper

Where does the new information about prices originate?Are there volatility spillovers between dairy markets?Did speculators contribute to rising volatility in the market?

8Slide9

Pricing Milk in the U.S. : 1. Government

9Slide10

Spot

market trades daily for 15 minutes each morning.No cash market for dry whey or milk.

10

Pricing Milk in the U.S. :

2. CME Cash MarketSlide11

Thin Slicing

11

Markets are very thin

USDA reports results of daily trading as well as weekly average

Prices for cheese used as benchmark in setting prices in direct transactions across the nationSlide12

12

Pricing Milk in the U.S. :

3. CME Futures MarketSlide13

Class III Milk Futures:

Comparing mid-October liquidity 2000-201113Slide14

Functions of the futures market: Price Discovery

14Slide15

Questions of interest

How do futures and cash market for cheese interact?Price discoveryVolatility spilloversImpact of speculation on dairy futures

15Slide16

A typical modeling approach

Test if cash and futures are stationaryIf yes: VARIf no: Co-integrationVolatility spillovers:If high-frequency: realized volatility/VARIf low-frequency: GARCH

Effects of speculationIf high-frequency: additional regressor in VARIf low-frequency: BEKK-X, EGARCH-X

16Slide17

VAR vs. co-integration

17

Case 1: Variables of interest are stationary (no persistent shocks)

Instruction: Build a vector autoregressive model

Case 2: Variables are non-stationary (some shocks are persistent)

Instruction: Build a co-integration modelSlide18

Data limitations

Cash market is thinClosing price may indicate unfilled bid/uncovered offerNo cash market for manufacturing grade milk or dry wheyFutures marketCheese futures market did not exist until 07/2010Data on speculative positions available only weekly

18Slide19

Implied Cheese Futures

19Slide20

Implied vs. observed cheese futures

20Slide21

21

Creating Nearby Futures Price SeriesSlide22

Unit root tests of cheese cash and futures time series

Augmented Dickey-Fuller (Said and Dickey, 1984)

Null: : (unit root present; no drift)2. Phillips-Perron

(1988):

Null: alpha=0,

rh

1

22Slide23

Unit Root Tests Results: Cash Cheese

23Slide24

Unit Root Tests Results: Cheese Futures

24Slide25

Devil is in the details: accounting for past lagged differenced futures

25Slide26

Unit Root Tests Results: Cheese Futures

26Slide27

Making sense of unit root results:

1. Economic TheoryCash price analysis based on production theory

Perfect competition: zero long-run economic profit for the marginal producerProfit margin will be a mean-reverting time series

If long-run industry average cost curve is flat

Permanent shifts in demand  temporary shifts to cash prices

Permanent changes in input prices  structural change

If supply is inelastic in short run  high persistency of shocks

If long-run AC curve is sloped

 Permanent shifts in demand  permanent shocks to cash price series

27Slide28

Making sense of unit root results:

1. Economic TheoryFutures price analysis based on finance theory

Efficient market prices in a single contract will be martingales if the marginal risk premium is zero;

submartingales

(downward biased) if marginal risk premium is positive

Supermartingales

(upward biased) if marginal risk premium is negative

- In any case: efficient futures prices will be

non-stationary

, i.e. all shocks to futures prices are permanent

28Slide29

Making sense of unit root results:

2. Time Series Modeling ExerciseWhat if there was a market in which cash price was indeed second-order stationaryIf there was a futures contract designed to cash-settle against such a spot price, what would be the characteristics of that time series?

For simplicity, assume no marginal risk premium

29Slide30

Making sense of unit root results:

2. Time Series Modeling Exercise

30Slide31

Making sense of unit root results:

2. Time Series Modeling Exercise - Results31

Martingale Property within each contract

Nearby series

not

a martingaleSlide32

Making sense of unit root results:

2. Time Series Modeling Exercise -What would Unit Root Tests Show?32

Cash Prices:

1) Null would likely be rejected

Futures prices:

2) for a single contract, null would likely not be rejected

3) Null more likely to be rejected for

n-

th

than for

n+1

nearby series

4) More obs. between rollover periods  null less likely to be rejected

(reducing data frequency increases likelihood of rejecting the null)Slide33

Unit Root Tests: Conclusions

Cash Cheese is mean revertingNearby cheese futures are nonlinearUnit-root processes within each contractMean-reverting at contract rollover

Next: How to model this?

33Slide34

Modeling information flows

Causality in meanSecond-order causality (causality in variance)

34Slide35

Second order non-causality

Granger non-causality: knowing the futures price does not help us predict cash (and vice versa).Second-order non-causality: knowing the futures price history may or may not help you predict the cash price level, but it does not influence the magnitude of cash price forecast conditional variance

Non-causality in variance: Granger non-causality and second-order non-causality combined

35Slide36

GARCH-BEKK and second-order non-causality

36Slide37

Adding speculators

The key problem is how to preserve positive definiteness of conditional variance matrixAdding another term?

Sign of the impact of additional regressor is restricted to be positive  but we must have flexibility!

37Slide38

GARCH-MEX

38Slide39

GARCH-MEX

39Slide40

Measuring “Adequacy” of Speculation

Based on Working (1960) – “Working’s T”The idea is that when hedgers are net long, long speculative position is not really ‘necessary’. But if it is there, it may “grease up” the market, or may be indicative of excessive speculation if T is too high.

So, if

40Slide41

Measuring “Adequacy” of Speculation

Likewise, if hedgers are net short, then only long speculative positions are needed to balance the market. Having long speculators may help, but too much of it may be “excessive”. So, if

Key assumption: how to treat unreportables.

41Slide42

Results: Information flows in mean

42Slide43

Results: Information flows in mean

43Slide44

Results: Information flows in mean

44Slide45

Results: Information flows in mean

45Slide46

Results: Information flows in mean

Conclusion: Using daily close prices at either daily or weekly frequency, using either nominal or log prices, and either control for heteroskedasticity or not – we always find that adjustment to spread between cash and futures is done in the cash market

46Slide47

Results: volatility spillovers

47

In a model where only GARCH-BEKK is added to error-correction model for mean, we find bi-directional volatility spillovers. Slide48

Results: Speculative Influence

48Slide49

Conclusions

Not likely that speculators increased volatility in dairy futures; if anything, speculative presence seems to be below what is deemed required for liquid market.GARCH-MEX has a potential for allowing flexible functional form, but restriction on correlation coefficient may flip the sign (and reduce the likelihood)

49Slide50

50